Corner Detection within a Multiscale Framework

We present a new multi scale method for corner detection. The proposed algorithm embodies an undecorated wavelet decomposition of the angulations signal of a shape contour to identify significant points on it. It detects peaks that persist through several scales from the correlation signal between scales of its non-orthogonal sub-band decompositions. These peaks correspond to high curvature points (HCPs). Furthermore, we compare the proposed method with others available in the literature, including the well-known curvature scale-space (CSS) method. The quantitative assessment of the algorithms is provided by some figures of merit (FOM) measures that indicate which method better detects the relevant points in terms of compaction and shape reconstruction.